Skip to content

Continuous Forge Tests - Unstable #330

Continuous Forge Tests - Unstable

Continuous Forge Tests - Unstable #330

Workflow file for this run

# Continuously run unstable forge tests against the latest main branch, to promote to stable.
name: Continuous Forge Tests - Unstable
permissions:
issues: write
pull-requests: write
contents: read
id-token: write
actions: write #required for workflow cancellation via check-aptos-core
on:
# Allow triggering manually
workflow_dispatch:
inputs:
IMAGE_TAG:
required: false
type: string
description: The docker image tag to test. This may be a git SHA1, or a tag like "<branch>_<git SHA1>". If not specified, Forge will find the latest build based on the git history (starting from GIT_SHA input)
GIT_SHA:
required: false
type: string
description: The git SHA1 to checkout. This affects the Forge test runner that is used. If not specified, the latest main will be used
pull_request:
paths:
- ".github/workflows/forge-unstable.yaml"
env:
AWS_ACCOUNT_NUM: ${{ secrets.ENV_ECR_AWS_ACCOUNT_NUM }}
AWS_ACCESS_KEY_ID: ${{ secrets.AWS_ACCESS_KEY_ID }}
AWS_SECRET_ACCESS_KEY: ${{ secrets.AWS_SECRET_ACCESS_KEY }}
AWS_REGION: us-west-2
IMAGE_TAG: ${{ inputs.IMAGE_TAG }} # this is only used for workflow_dispatch, otherwise defaults to empty
jobs:
# This job determines the image tag and branch to test, and passes them to the other jobs
# NOTE: this may be better as a separate workflow as the logic is quite complex but generalizable
determine-test-metadata:
runs-on: ubuntu-latest
outputs:
IMAGE_TAG: ${{ steps.get-docker-image-tag.outputs.IMAGE_TAG }}
BRANCH: ${{ steps.determine-test-branch.outputs.BRANCH }}
steps:
- uses: actions/checkout@v4
- name: Determine branch based on cadence
id: determine-test-branch
run: |
if [[ "${{ github.event_name }}" == "schedule" ]]; then
echo "Unknown schedule: ${{ github.event.schedule }}"
exit 1
else
echo "Using GIT_SHA"
# on workflow_dispatch, this will simply use the inputs.GIT_SHA given (or the default)
# on pull_request, this will default to null and the following "checkout" step will use the PR's base branch
echo "BRANCH=${{ inputs.GIT_SHA }}" >> $GITHUB_OUTPUT
fi
- uses: aptos-labs/aptos-core/.github/actions/check-aptos-core@main
with:
cancel-workflow: ${{ github.event_name == 'schedule' }} # Cancel the workflow if it is scheduled on a fork
# actions/get-latest-docker-image-tag requires docker utilities and having authenticated to internal docker image registries
- uses: aptos-labs/aptos-core/.github/actions/docker-setup@main
id: docker-setup
with:
GCP_WORKLOAD_IDENTITY_PROVIDER: ${{ secrets.GCP_WORKLOAD_IDENTITY_PROVIDER }}
GCP_SERVICE_ACCOUNT_EMAIL: ${{ secrets.GCP_SERVICE_ACCOUNT_EMAIL }}
EXPORT_GCP_PROJECT_VARIABLES: "false"
AWS_ACCESS_KEY_ID: ${{ secrets.AWS_ACCESS_KEY_ID }}
AWS_SECRET_ACCESS_KEY: ${{ secrets.AWS_SECRET_ACCESS_KEY }}
AWS_DOCKER_ARTIFACT_REPO: ${{ secrets.AWS_DOCKER_ARTIFACT_REPO }}
GIT_CREDENTIALS: ${{ secrets.GIT_CREDENTIALS }}
- uses: aptos-labs/aptos-core/.github/actions/get-latest-docker-image-tag@main
id: get-docker-image-tag
with:
branch: ${{ steps.determine-test-branch.outputs.BRANCH }}
variants: "failpoints performance"
- name: Write summary
run: |
IMAGE_TAG=${{ steps.get-docker-image-tag.outputs.IMAGE_TAG }}
BRANCH=${{ steps.determine-test-branch.outputs.BRANCH }}
if [ -n "${BRANCH}" ]; then
echo "BRANCH: [${BRANCH}](https://github.com/${{ github.repository }}/tree/${BRANCH})" >> $GITHUB_STEP_SUMMARY
fi
echo "IMAGE_TAG: [${IMAGE_TAG}](https://github.com/${{ github.repository }}/commit/${IMAGE_TAG})" >> $GITHUB_STEP_SUMMARY
forge-continuous:
if: ${{ github.event_name != 'pull_request' }}
needs: determine-test-metadata
uses: ./.github/workflows/workflow-run-forge.yaml
secrets: inherit
with:
IMAGE_TAG: ${{ needs.determine-test-metadata.outputs.IMAGE_TAG }}
# GCP cluster
FORGE_CLUSTER_NAME: aptos-forge-1
COMMENT_HEADER: forge-continuous
# This test suite is configured using the forge.py config test command
FORGE_TEST_SUITE: continuous
run-forge-state-sync-slow-processing-catching-up-test:
if: ${{ github.event_name != 'pull_request' && always() }}
needs: [ determine-test-metadata, forge-continuous ] # Only run after the previous job completes
uses: aptos-labs/aptos-core/.github/workflows/workflow-run-forge.yaml@main
secrets: inherit
with:
IMAGE_TAG: ${{ needs.determine-test-metadata.outputs.IMAGE_TAG }}
# GCP cluster
FORGE_CLUSTER_NAME: aptos-forge-1
FORGE_NAMESPACE: forge-state-sync-slow-processing-catching-up-test-${{ needs.determine-test-metadata.outputs.IMAGE_TAG }}
FORGE_RUNNER_DURATION_SECS: 900 # Run for 15 minutes
FORGE_TEST_SUITE: state_sync_slow_processing_catching_up
POST_TO_SLACK: true
FORGE_ENABLE_FAILPOINTS: true
run-forge-twin-validator-test:
if: ${{ github.event_name != 'pull_request' && always() }}
needs: [ determine-test-metadata, run-forge-state-sync-slow-processing-catching-up-test ] # Only run after the previous job completes
uses: aptos-labs/aptos-core/.github/workflows/workflow-run-forge.yaml@main
secrets: inherit
with:
IMAGE_TAG: ${{ needs.determine-test-metadata.outputs.IMAGE_TAG }}
# GCP cluster
FORGE_CLUSTER_NAME: aptos-forge-1
FORGE_NAMESPACE: forge-twin-validator-${{ needs.determine-test-metadata.outputs.IMAGE_TAG }}
FORGE_RUNNER_DURATION_SECS: 900 # Run for 15 minutes
FORGE_TEST_SUITE: twin_validator_test
POST_TO_SLACK: true
run-forge-state-sync-failures-catching-up-test:
if: ${{ github.event_name != 'pull_request' && always() }}
needs: [ determine-test-metadata, run-forge-twin-validator-test ] # Only run after the previous job completes
uses: aptos-labs/aptos-core/.github/workflows/workflow-run-forge.yaml@main
secrets: inherit
with:
IMAGE_TAG: ${{ needs.determine-test-metadata.outputs.IMAGE_TAG }}
FORGE_NAMESPACE: forge-state-sync-failures-catching-up-test-${{ needs.determine-test-metadata.outputs.IMAGE_TAG }}
# GCP cluster
FORGE_CLUSTER_NAME: aptos-forge-1
FORGE_RUNNER_DURATION_SECS: 900 # Run for 15 minutes
FORGE_TEST_SUITE: state_sync_failures_catching_up
FORGE_ENABLE_FAILPOINTS: true
POST_TO_SLACK: ${{ needs.determine-test-metadata.outputs.BRANCH == 'main' }} # only post to slack on main branch
run-forge-validator-reboot-stress-test:
if: ${{ github.event_name != 'pull_request' && always() }}
needs: [ determine-test-metadata, run-forge-state-sync-failures-catching-up-test ] # Only run after the previous job completes
uses: aptos-labs/aptos-core/.github/workflows/workflow-run-forge.yaml@main
secrets: inherit
with:
IMAGE_TAG: ${{ needs.determine-test-metadata.outputs.IMAGE_TAG }}
# GCP cluster
FORGE_CLUSTER_NAME: aptos-forge-1
FORGE_NAMESPACE: forge-validator-reboot-stress-${{ needs.determine-test-metadata.outputs.IMAGE_TAG }}
FORGE_RUNNER_DURATION_SECS: 2400 # Run for 40 minutes
FORGE_TEST_SUITE: validator_reboot_stress_test
POST_TO_SLACK: true
run-forge-haproxy:
if: ${{ github.event_name != 'pull_request' && always() }}
needs: [ determine-test-metadata, run-forge-validator-reboot-stress-test ] # Only run after the previous job completes
uses: aptos-labs/aptos-core/.github/workflows/workflow-run-forge.yaml@main
secrets: inherit
with:
IMAGE_TAG: ${{ needs.determine-test-metadata.outputs.IMAGE_TAG }}
# GCP cluster
FORGE_CLUSTER_NAME: aptos-forge-1
FORGE_NAMESPACE: forge-haproxy-${{ needs.determine-test-metadata.outputs.BRANCH_HASH }}
FORGE_RUNNER_DURATION_SECS: 600 # Run for 10 minutes
FORGE_ENABLE_HAPROXY: true
FORGE_TEST_SUITE: realistic_env_max_load
POST_TO_SLACK: true